Implicit Hamiltonian Monte Carlo for Sampling Multiscale Distributions

Hamiltonian Monte Carlo (HMC) has been widely adopted in the statistics community because of its ability to sample high-dimensional distributions much more efficiently than other Metropolis-based methods. Despite this, HMC often performs sub-optimally on distributions with high correlations or marginal variances on multiple scales because the resulting stiffness forces the leapfrog integrator in HMC to take an unreasonably small stepsize. We provide intuition as well as a formal analysis showing how these multiscale distributions limit the stepsize of leapfrog and we show how the implicit midpoint method can be used, together with Newton-Krylov iteration, to circumvent this limitation and achieve major efficiency gains. Furthermore, we offer practical guidelines for when to choose between implicit midpoint and leapfrog and what stepsize to use for each method, depending on the distribution being sampled. Unlike previous modifications to HMC, our method is generally applicable to highly non-Gaussian distributions exhibiting multiple scales. We illustrate how our method can provide a dramatic speedup over leapfrog in the context of the No-U-Turn sampler (NUTS) applied to several examples.

[1]  Andrew Gelman,et al.  The No-U-turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo , 2011, J. Mach. Learn. Res..

[2]  David B. Dunson,et al.  Bayesian Data Analysis , 2010 .

[3]  Y. Saad,et al.  GMRES: a generalized minimal residual algorithm for solving nonsymmetric linear systems , 1986 .

[4]  Uri M. Ascher,et al.  Computer methods for ordinary differential equations and differential-algebraic equations , 1998 .

[5]  Carl D. Meyer,et al.  Matrix Analysis and Applied Linear Algebra , 2000 .

[6]  Radford M. Neal MCMC Using Hamiltonian Dynamics , 2011, 1206.1901.

[7]  E. Hairer,et al.  Geometric Numerical Integration: Structure Preserving Algorithms for Ordinary Differential Equations , 2004 .

[8]  Homer F. Walker,et al.  Choosing the Forcing Terms in an Inexact Newton Method , 1996, SIAM J. Sci. Comput..

[9]  Uri M. Ascher,et al.  On Some Difficulties in Integrating Highly Oscillatory Hamiltonian Systems , 1999, Computational Molecular Dynamics.

[10]  A. Kennedy,et al.  Hybrid Monte Carlo , 1988 .

[11]  Hideyuki Suzuki,et al.  Hamiltonian Monte Carlo with explicit, reversible, and volume-preserving adaptive step size control , 2017, JSIAM Lett..

[12]  N. Metropolis,et al.  Equation of State Calculations by Fast Computing Machines , 1953, Resonance.

[13]  D. Keyes,et al.  Jacobian-free Newton-Krylov methods: a survey of approaches and applications , 2004 .

[14]  Michael Betancourt,et al.  A General Metric for Riemannian Manifold Hamiltonian Monte Carlo , 2012, GSI.

[15]  Michael Betancourt,et al.  A Conceptual Introduction to Hamiltonian Monte Carlo , 2017, 1701.02434.

[16]  Wolfram Burgard,et al.  Robotics: Science and Systems XV , 2010 .

[17]  J. M. Sanz-Serna,et al.  Optimal tuning of the hybrid Monte Carlo algorithm , 2010, 1001.4460.

[18]  Babak Shahbaba,et al.  Split Hamiltonian Monte Carlo , 2011, Stat. Comput..

[19]  Jiqiang Guo,et al.  Stan: A Probabilistic Programming Language. , 2017, Journal of statistical software.

[20]  Justin Solomon,et al.  Exponential Integration for Hamiltonian Monte Carlo , 2015, ICML.

[21]  Gregory S. Chirikjian,et al.  The Banana Distribution is Gaussian: A Localization Study with Exponential Coordinates , 2012, Robotics: Science and Systems.